name = "MiniMax-M3"
description = "MiniMax multimodal model for long-context coding, perception, and agent planning"
family = "minimax"
release_date = "2026-06-01"
last_updated = "2026-06-01"
attachment = true
reasoning = true
temperature = true
tool_call = true
open_weights = true

[limit]
context = 512_000
output = 128_000

[modalities]
input = ["text", "image", "video"]
output = ["text"]

[[weights]]
label = "Hugging Face"
url = "https://huggingface.co/MiniMaxAI/MiniMax-M3"

[[benchmarks]]
name = "SWE-Bench Verified"
score = 80.5
metric = "resolved"
harness = "Claude Code"
source = "https://www.minimax.io/blog/minimax-m3"
date = "2026-06-01"

[[benchmarks]]
name = "SWE-Bench Pro"
score = 59.0
metric = "resolve rate"
harness = "Claude Code"
source = "https://www.minimax.io/blog/minimax-m3"
date = "2026-06-01"

[[benchmarks]]
name = "Terminal-Bench"
score = 66.0
metric = "success rate"
version = "2.1"
source = "https://www.minimax.io/blog/minimax-m3"
date = "2026-06-01"

[[benchmarks]]
name = "BrowseComp"
score = 83.52
metric = "accuracy"
source = "https://www.minimax.io/blog/minimax-m3"
date = "2026-06-01"

[[benchmarks]]
name = "MCP Atlas"
score = 74.2
metric = "success rate"
source = "https://www.minimax.io/blog/minimax-m3"
date = "2026-06-01"

[[benchmarks]]
name = "OSWorld-Verified"
score = 70.06
metric = "success rate"
source = "https://www.minimax.io/blog/minimax-m3"
date = "2026-06-01"